JOIG A0005: Optic Disc and Optic Cup Segmentation on Retinal Image based on Multimap Localization and U-Net Convolutional Neural Network

Abstract

Glaucoma is an eye disease that is caused by an increase in the intraocular pressure. Excessive intraocular pressure could damage optic nerves and lead to vision loss. Early detection is crucial to treat the patient as fast as possible to prevent irreversible blindness. Glaucoma diagnosis can be supported by a computer-aided detection system that provides an automated measurement of the physical dimensions of the optic disc and the optic cup from retinal images. The cup to disc ratio (CDR) of these measurements indicates the risk of developing glaucoma. CDR calculation relies on accurate optic disc localization and optic disc/cup segmentation. In this research, we developed an unsupervised optic disc localization and optic disc/cup semantic segmentation using U-Net which is optimized for retinal images from the Drishti-GS and Refuge datasets. 99.99 ± 0.12% of the optic disc pixels were correctly included in the regions of interest by the proposed localization method. Further semantic segmentation on the regions of interest yields F-scores of 0.935 ± 0.031 (Drishti-GS) and 0.950 ± 0.028 (Refuge) for optic disc and 0.832 ± 0.071 (Drishti-GS) and 0.871 ± 0.061 (Refuge) for optic cup. These results are comparable to the state-of-the-art methods.